Pith. sign in

REVIEW 3 major objections 5 minor 42 references

Pose-prompted video segmentation of hands and body raises isolated sign language recognition accuracy to 72.8% on the ChaLearn249 benchmark.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-04 17:34 UTC pith:6LFBVOCV

load-bearing objection A clean, well-explained application of SAM 2 to isolate sign-language body/hand regions, with a real accuracy gain on ChaLearn249 IsoGD but the central mechanism—segmentation quality—is never actually measured. the 3 major comments →

arxiv 2509.10710 v1 pith:6LFBVOCV submitted 2025-09-12 cs.CV

SegSLR: Promptable Video Segmentation for Isolated Sign Language Recognition

Sven Schreiber , Noha Sarhan , Simone Frintrop , Christian Wilms This is my paper
classification cs.CV
keywords sign language recognitionaction recognitionsegmentationpromptable video segmentationSAM 2pose estimationI3D CNNChaLearn249 IsoGD
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper argues that isolated sign language recognition can be improved substantially by using pose estimation only as a prompt for a pretrained video segmentation model, rather than as a separate input or as bounding-box crops. SegSLR feeds RTMW keypoints into SAM 2 to obtain temporally consistent, pixel-accurate masks of the signer's body and hands, then classifies masked RGB frames and segmentation logits with I3D CNNs. On the ChaLearn249 IsoGD dataset the system reaches 71.30% validation and 72.76% test accuracy, outperforming prior methods by up to 4.17 percentage points. The gains come from retaining hand shape, finger detail, and body posture that bounding-box representations discard.

Core claim

SegSLR establishes that promptable zero-shot video segmentation can serve as the bridge between RGB and pose modalities for isolated sign language recognition. The core finding is that using pose keypoints as positive and negative point prompts for SAM 2 produces masklets of the hands and body that preserve detailed shape information throughout the video, and that feeding these masks (both as masked RGB frames and as raw logits) through standard I3D CNNs yields large accuracy gains: +9.21 validation / +8.32 test points over the RGB+flow baseline. The ablation shows each segmentation stream contributes independently, with hands contributing the largest improvement.

What carries the argument

The central mechanism is the promptable video segmentation module: RTMW estimates 116 whole-body keypoints; a best frame is selected by maximizing keypoint confidence and minimizing hand–face bounding-box overlap; from that frame, hand and body keypoint subsets are turned into positive/negative point prompts for SAM 2, which propagates masklets bidirectionally through the video. These masks are used in four I3D streams: masked RGB of body, body logits, masked RGB of hands, hand logits, combined by score-level fusion.

Load-bearing premise

The load-bearing premise is that SAM 2, prompted by RTMW keypoints, produces accurately segmented and temporally consistent hand and body masks that preserve the finger- and orientation-level details needed for ISLR; the paper reports downstream accuracy but does not measure mask quality directly.

What would settle it

Run a masked-IoU evaluation of SAM 2's hand and body segmentations against manual annotations on a sample of ChaLearn249 frames; if IoU is low or inconsistent on hands-in-front-of-face or fast-motion clips, the attribution of the accuracy gain to segmentation quality would be falsified. Alternatively, replace SAM 2 masks with tight bounding-box crops of the same regions: if accuracy does not drop, the claim that pixel-precise masks matter is weakened.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

If this is right

  • If correct, hybrid RGB+pose ISLR no longer needs skeleton graphs or bounding-box crops; a general video segmentation backbone can supply the shape and orientation cues sign language depends on.
  • The approach is zero-shot in both segmentation and pose, so it should transfer to other sign languages or datasets without retraining those components.
  • The ablation indicates body and hand streams contribute independently, so the design can be extended to face or other non-manual markers by adding analogous prompts.
  • The use of both masked frames and logits suggests that per-pixel confidence carries useful location and context information, not just the binary mask.
  • On the benchmark, SegSLR outperforms the strongest published baseline by 4.17 validation points and 1.85 test points, indicating a real but narrower edge on the test split.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Beyond the paper: because no segmentation-quality metric (e.g., IoU with manual masks) is reported, the causal claim that mask fidelity drives the gain is not directly tested; a dedicated mask-quality study would close this gap.
  • Beyond the paper: the method depends on RTMW keypoint detection, and when keypoints are missing the prompts are dropped; a prompt-failure analysis or fallback to box prompts could reveal robustness limits.
  • Beyond the paper: the independent I3D streams and score fusion make it straightforward to add a pose-based GCN stream in a late-fusion ensemble, which the paper does not test.
  • Beyond the paper: the best-frame selection heuristic is hand-crafted; learning the prompt frame or using multiple prompt frames is a natural extension that could improve robustness on challenging videos.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes SegSLR, a multi-stream isolated sign language recognition (ISLR) system that combines RGB and pose information through promptable video segmentation. RTMW pose keypoints are used to generate point prompts for SAM 2, which produces temporally consistent masklets of the signer's body and hands. These masklets are consumed by four I3D-CNN streams (masked RGB and segmentation logits for body and hands), alongside existing RGB and optical-flow streams, with score-level fusion. On ChaLearn249 IsoGD, the paper reports 71.30% validation and 72.76% test accuracy (median of five runs; mean 71.39, std 0.41), outperforming its I3D-SLR baseline and several prior ISLR methods. Ablations show monotonic gains when adding body and hand segmentation streams, and Table 3 compares SAM 2 with SAM and Mask R-CNN.

Significance. If the results hold, SegSLR provides a clean way to inject pose information into RGB-based ISLR without training a separate pose model, leveraging frozen foundation models. The paper's strengths include: the main result is reported as a median with mean and standard deviation over five training runs; evaluation is on a held-out validation/test split; SAM 2 and RTMW are used off-the-shelf, making the pipeline reproducible in principle; and the ablation path in Table 2 is monotonic. However, the significance is currently tempered by three issues: (1) the causal mechanism named in the abstract and conclusion—that SAM 2 masks preserve hand shape, orientation, and finger detail better than bounding boxes—is not directly measured; (2) the ablations in Table 2 lack error bars; and (3) the 'state-of-the-art' claim relies on a single dataset and a comparison set that is largely older and partly from the authors' own prior work. The central accuracy claim is plausible, but the explanatory claim needs additional evidence or explicit softening.

major comments (3)
  1. [Sec. 5.2, Table 3; Secs. 1/2] The paper's motivating contrast is between high-quality SAM 2 masklets and imprecise bounding-box crops (Sec. 1 and Sec. 2). This contrast is never tested within the proposed architecture. Table 3 varies only the segmentation method (Mask R-CNN, SAM, SAM 2) while keeping the stream design fixed; it does not include a stream that uses keypoint-derived bounding boxes or binary box masks. In addition, Table 2 adds segmentation streams without a control that adds an equal number of non-segmentation streams. The reported gains are therefore compatible with a coarse attention/focusing effect or with score-fusion ensembling, not only with the claimed preservation of hand shape, orientation, and finger detail. Please add an ablation replacing SAM 2 masklets with pose-derived boxes (or dilated keypoint masks) in an otherwise identical six-stream configuration, and, if feasible, report mask qualit
  2. [Sec. 5.2, Table 2] Ablation rows lack error bars, making the incremental gains difficult to interpret relative to the reported seed noise.
  3. [Sec. 5.1, Table 1] The state-of-the-art claim is broader than the evidence.
minor comments (5)
  1. [Table 1 vs. Table 2] The baseline I3D-SLR test accuracy is 64.44 in Table 1 but 64.42 in the first row of Table 2. Please make the numbers consistent.
  2. [Sec. 4.2] The best-frame scoring function is defined as a product of three normalized terms, but no sensitivity analysis is provided for the choice of terms or their weighting. A sentence explaining the behavior would help, since hand-to-face overlap and keypoint confidence could conflict.
  3. [Sec. 4.1] The selection of positive/negative keypoint subsets is manual and not ablated. It would be useful to know how sensitive downstream accuracy is to, for example, adding the ignored hand keypoints or omitting the negative face prompts.
  4. [Throughout] The term 'state-of-the-art' is used repeatedly; consider using 'compared baselines' for precision. Also, 'up to 4.17%' in the contributions should specify whether this is an absolute percentage-point difference (it is, relative to TD-SLR on validation).
  5. [References] Reference [31] contains a typo: 'Fritnrop' should be 'Frintrop.'

Circularity Check

0 steps flagged

No circularity: SegSLR's improvement is measured on held-out data; ablations and baselines are independent comparisons, not inputs to the method.

full rationale

The paper's central claim is an empirical accuracy comparison on the ChaLearn249 IsoGD validation and test splits, with SegSLR trained five times on the training set and the median validation model reported (Sec. 5). The ablations (Tables 2 and 3) add segmentation streams stepwise while keeping the I3D training protocol fixed, so each increment is attributable to the added component rather than to a fitted parameter. Baselines [28,29,31,32] are the authors' prior published works, but they are used only as comparison points with independently reported numbers; they are not parameters, constraints, or justifications in SegSLR's construction. No equation in the paper defines the output in terms of the input, no parameter is fitted to the test set, and no claim is justified by a self-citation chain. The lack of a direct mask-quality metric (e.g., IoU) is a limitation of the causal interpretation that segmentation quality drives the gain, but this is a correctness/evidence concern, not circularity. The derivation chain is self-contained: RTMW keypoints prompt SAM 2, the resulting masks and logits are fed to I3D streams, and the final accuracy is externally measured against held-out data.

Axiom & Free-Parameter Ledger

3 free parameters · 4 axioms · 0 invented entities

The central result is empirical. It assumes the pre-trained components (RTMW pose estimation, SAM 2 segmentation, ImageNet/Kinetics I3D) transfer to sign language data, and that the hand-designed prompt and best-frame selection heuristics are reasonable. There are no fitted parameters used to derive a prediction; the reported numbers are measured on held-out splits.

free parameters (3)
  • Best-frame selection scoring components = equal-weight product of normalized average keypoint confidence, keypoint bounding-box area, and 1 minus hand-face overla
    Sec. 4.2; chosen by hand, no sensitivity analysis or learned weights.
  • Positive/negative point prompt keypoint subsets = body: all RTMW keypoints except fine hand/face; hands: first joint per finger positives; negatives: major body and face
    Sec. 4.1; hand-selected subsets that directly determine SAM 2 outputs; no comparison of alternative prompt sets.
  • Training setup (frames, crop, batch, patience) = 40 frames, 224x224 central crop, batch size 4, early stopping patience 3
    Sec. 4.5; standard I3D training choices, but no sensitivity analysis.
axioms (4)
  • domain assumption SAM 2 pre-trained on SA-1B and SA-V segments signer body and hands from RTMW point prompts in a zero-shot manner with sufficient accuracy and temporal consistency.
    Assumed in Secs. 4.3-4.5; no fine-tuning because no ISLR segmentation annotations; only qualitative evidence in Fig. 6.
  • domain assumption RTMW pose estimation yields accurate 116-keypoint whole-body poses for sign language videos.
    Sec. 4.1; RTMW is pre-trained on 14 datasets but not evaluated on signers; errors propagate to prompts.
  • domain assumption Score-level fusion of six independently trained I3D streams is an effective combination strategy.
    Sec. 4.4-5.2; empirically supported by ablations, but no theoretical or statistical justification.
  • domain assumption ChaLearn249 IsoGD is a valid benchmark for evaluating ISLR accuracy.
    Sec. 5; the dataset is assumed to represent the target task and its labels are assumed correct.

pith-pipeline@v1.3.0-alltime-deepseek · 9933 in / 9974 out tokens · 109327 ms · 2026-08-04T17:34:46.822232+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of SegSLR: Promptable Video Segmentation for Isolated Sign Language Recognition." pith.science (2026). https://pith.science/paper/6LFBVOCV

@misc{pith2026250910710,
  author       = {Pith},
  title        = {Pith review of: SegSLR: Promptable Video Segmentation for Isolated Sign Language Recognition},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6LFBVOCV}},
  note         = {Machine review of arXiv:2509.10710}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Isolated Sign Language Recognition (ISLR) approaches primarily rely on RGB data or signer pose information. However, combining these modalities often results in the loss of crucial details, such as hand shape and orientation, due to imprecise representations like bounding boxes. Therefore, we propose the ISLR system SegSLR, which combines RGB and pose information through promptable zero-shot video segmentation. Given the rough localization of the hands and the signer's body from pose information, we segment the respective parts through the video to maintain all relevant shape information. Subsequently, the segmentations focus the processing of the RGB data on the most relevant body parts for ISLR. This effectively combines RGB and pose information. Our evaluation on the complex ChaLearn249 IsoGD dataset shows that SegSLR outperforms state-of-the-art methods. Furthermore, ablation studies indicate that SegSLR strongly benefits from focusing on the signer's body and hands, justifying our design choices.

Figures

Figures reproduced from arXiv: 2509.10710 by Christian Wilms, Noha Sarhan, Simone Frintrop, Sven Schreiber.

Figure 1
Figure 1. Figure 1: Idea of the proposed SegSLR system: We combine RGB information (first [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of our proposed SegSLR system. Based on RGB frames of a [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Overview of RTMW [15] keypoints for prompting SAM 2 to segment the [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Detailed view of our promptable video segmentation module for hands [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Examples of the body segmentation, body logits, hands segmentation, and [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Qualitative body (green masks) and hands (blue masks) segmentation [PITH_FULL_IMAGE:figures/full_fig_p011_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Comparison of segmentations of the signer’s body with SAM (upper [PITH_FULL_IMAGE:figures/full_fig_p012_7.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

42 extracted references · 4 linked inside Pith

  1. [1]

    Gallaudet University Press (1991)

    Baker-Shenk, C.L., Cokely, D.: American Sign Language: A teacher’s resource text on grammar and culture. Gallaudet University Press (1991)

    1991

  2. [2]

    In: Conference on Computer Vision and Pattern Recognition (2017)

    Carreira, J., Zisserman, A.: Quo vadis, action recognition? a new model and the ki- netics dataset. In: Conference on Computer Vision and Pattern Recognition (2017)

    2017

  3. [3]

    In: International Conference on Computer Vision (2023)

    Cheng, H.K., Oh, S.W., Price, B., Schwing, A., Lee, J.Y.: Tracking anything with decoupled video segmentation. In: International Conference on Computer Vision (2023)

    2023

  4. [4]

    In: Computer Vision and Pattern Recognition Workshop (2021)

    De Coster, M., Van Herreweghe, M., Dambre, J.: Isolated sign recognition from RGB video using pose flow and self-attention. In: Computer Vision and Pattern Recognition Workshop (2021)

    2021

  5. [5]

    arXiv preprint arXiv:1810.04805 (2018)

    Devlin, J.: BERT pre-training of deep bidirectional transformers for language un- derstanding. arXiv preprint arXiv:1810.04805 (2018)

    Pith/arXiv arXiv 2018

  6. [6]

    Transactions on Multimedia Computing, Communi- cations, and Applications (2018)

    Duan, J., Wan, J., Zhou, S., Guo, X., Li, S.Z.: A unified framework for multi-modal isolated gesture recognition. Transactions on Multimedia Computing, Communi- cations, and Applications (2018)

    2018

  7. [7]

    In: European Conference on Computer Vision Workshop (2020)

    Gökçe, Ç., Özdemir, O., Kındıroğlu, A.A., Akarun, L.: Score-level multi cue fu- sion for sign language recognition. In: European Conference on Computer Vision Workshop (2020)

    2020

  8. [8]

    In: Computer Vision and Pat- tern Recognition Workshop (2021)

    Gruber, I., Krnoul, Z., Hrúz, M., Kanis, J., Bohacek, M.: Mutual support of data modalities in the task of sign language recognition. In: Computer Vision and Pat- tern Recognition Workshop (2021)

    2021

  9. [9]

    In: Advances in Neural Information Processing Sys- tems (2024)

    He, C., Li, K., Zhang, Y., Xu, G., Tang, L., Zhang, Y., Guo, Z., Li, X.: Weakly- supervised concealed object segmentation with SAM-based pseudo labeling and multi-scale feature grouping. In: Advances in Neural Information Processing Sys- tems (2024)

    2024

  10. [10]

    In: Conference on Computer Vision and Pattern Recogni- tion (2022)

    He, K., Chen, X., Xie, S., Li, Y., Dollár, P., Girshick, R.: Masked autoencoders are scalable vision learners. In: Conference on Computer Vision and Pattern Recogni- tion (2022)

    2022

  11. [11]

    In: International Conference on Computer Vision (2017)

    He, K., Gkioxari, G., Dollar, P., Girshick, R.: Mask R-CNN. In: International Conference on Computer Vision (2017)

    2017

  12. [12]

    In: Winter Con- ference on Applications of Computer Vision (2021)

    Hosain, A.A., Santhalingam, P.S., Pathak, P., Rangwala, H., Kosecka, J.: Hand pose guided 3D pooling for word-level sign language recognition. In: Winter Con- ference on Applications of Computer Vision (2021)

    2021

  13. [13]

    In: International Con- ference on Computer Vision (2021)

    Hu, H., Zhao, W., Zhou, W., Wang, Y., Li, H.: SignBERT: Pre-training of hand- model-aware representation for sign language recognition. In: International Con- ference on Computer Vision (2021)

    2021

  14. [14]

    In: Computer Vision and Pattern Recognition Workshop (2021)

    Jiang, S., Sun, B., Wang, L., Bai, Y., Li, K., Fu, Y.: Skeleton aware multi-modal sign language recognition. In: Computer Vision and Pattern Recognition Workshop (2021)

    2021

  15. [15]

    arXiv preprint arXiv:2407.08634 (2024)

    Jiang, T., Xie, X., Li, Y.: RTMW: Real-time multi-person 2d and 3d whole-body pose estimation. arXiv preprint arXiv:2407.08634 (2024)

    Pith/arXiv arXiv 2024

  16. [16]

    In: International Conference on Computer Vision (2023)

    Kirillov, A., Mintun, E., Ravi, N., Mao, H., Rolland, C., Gustafson, L., Xiao, T., Whitehead, S., Berg, A.C., Lo, W.Y., et al.: Segment anything. In: International Conference on Computer Vision (2023)

    2023

  17. [17]

    In: Computer Vision and Pattern Recognition Workshop (2018) 14 Sven Schreiber, Noha Sarhan, Simone Frintrop, and Christian Wilms

    Kopuklu, O., Kose, N., Rigoll, G.: Motion fused frames: Data level fusion strat- egy for hand gesture recognition. In: Computer Vision and Pattern Recognition Workshop (2018) 14 Sven Schreiber, Noha Sarhan, Simone Frintrop, and Christian Wilms

    2018

  18. [18]

    In: International Conference on Computer Vision (2023)

    Lee, T., Oh, Y., Lee, K.M.: Human part-wise 3d motion context learning for sign language recognition. In: International Conference on Computer Vision (2023)

    2023

  19. [19]

    In: International Conference on Image Processing (2018)

    Li, B., Li, W., Tang, Y., Hu, J.F., Zheng, W.S.: GL-PAM RGB-D gesture recog- nition. In: International Conference on Image Processing (2018)

    2018

  20. [20]

    In: Computer Vision and Pattern Recognition (2020)

    Li, D., Yu, X., Xu, C., Petersson, L., Li, H.: Transferring cross-domain knowledge for video sign language recognition. In: Computer Vision and Pattern Recognition (2020)

    2020

  21. [21]

    Multimedia Tools and Applications78(2019)

    Lim, K.M., Tan, A.W.C., Lee, C.P., Tan, S.C.: Isolated sign language recogni- tion using convolutional neural network hand modelling and hand energy image. Multimedia Tools and Applications78(2019)

    2019

  22. [22]

    Transactions on Multimedia26(2024)

    Lin, K., Wang, X., Zhu, L., Zhang, B., Yang, Y.: SKIM: Skeleton-based isolated sign language recognition with part mixing. Transactions on Multimedia26(2024)

    2024

  23. [23]

    In: European Conference on Computer Vision (2014)

    Lin, T.Y., Maire, M., Belongie, S., Hays, J., Perona, P., Ramanan, D., Dollár, P., Zitnick, C.L.: Microsoft COCO: Common objects in context. In: European Conference on Computer Vision (2014)

    2014

  24. [24]

    Nature Communications15(1) (2024)

    Ma, J., He, Y., Li, F., Han, L., You, C., Wang, B.: Segment anything in medical images. Nature Communications15(1) (2024)

    2024

  25. [25]

    Expert Systems with Applications (2021)

    Rastgoo, R., Kiani, K., Escalera, S.: Sign language recognition: A deep survey. Expert Systems with Applications (2021)

    2021

  26. [26]

    arXiv preprint arXiv:2408.00714 (2024)

    Ravi, N., Gabeur, V., Hu, Y.T., Hu, R., Ryali, C., Ma, T., Khedr, H., Rädle, R., Rolland, C., Gustafson, L., et al.: Sam 2: Segment anything in images and videos. arXiv preprint arXiv:2408.00714 (2024)

    Pith/arXiv arXiv 2024

  27. [27]

    International Journal of Computer Vision115(2015)

    Russakovsky, O., Deng, J., Su, H., Krause, J., Satheesh, S., Ma, S., Huang, Z., Karpathy, A., Khosla, A., Bernstein, M., et al.: Imagenet large scale visual recog- nition challenge. International Journal of Computer Vision115(2015)

    2015

  28. [28]

    In: International Conference on Image Processing (2020)

    Sarhan, N., Frintrop, S.: Transfer learning of videos: From action recognition to sign language recognition. In: International Conference on Image Processing (2020)

    2020

  29. [29]

    In: International Conference on Automatic Face and Gesture Recog- nition (2021)

    Sarhan, N., Frintrop, S.: Sign, attend and tell: Spatial attention for sign language recognition. In: International Conference on Automatic Face and Gesture Recog- nition (2021)

    2021

  30. [30]

    In: International Conference on Computer Vision Workshop (2023)

    Sarhan, N., Frintrop, S.: Unraveling a decade: A comprehensive survey on iso- lated sign language recognition. In: International Conference on Computer Vision Workshop (2023)

    2023

  31. [31]

    In: International Conference on Computer Vision Systems (2023)

    Sarhan, N., Willruth, J.M., Fritnrop, S.: Pseudodepth-slr: Generating depth data for sign language recognition. In: International Conference on Computer Vision Systems (2023)

    2023

  32. [32]

    Sarhan, N., Wilms, C., Closius, V., Brefeld, U., Frintrop, S.: Hands in focus: Sign languagerecognitionviatop-downattention.In:InternationalConferenceonImage Processing (2023)

    2023

  33. [33]

    In: Winter Conference on Applications of Computer Vision (2021)

    Tunga, A., Nuthalapati, S.V., Wachs, J.: Pose-based sign language recognition using GCN and BERT. In: Winter Conference on Applications of Computer Vision (2021)

    2021

  34. [34]

    Wan, J., Zhao, Y., Zhou, S., Guyon, I., Escalera, S., Li, S.Z.: Chalearn looking at peoplergb-d isolatedand continuousdatasets forgesture recognition.In: Computer Vision and Pattern Recognition Workshop (2016)

    2016

  35. [35]

    In: European Conference on Computer Vision (2024)

    Wilms, C., Rolff, T., Hillemann, M., Johanson, R., Frintrop, S.: SOS: Segment ob- ject system for open-world instance segmentation with object priors. In: European Conference on Computer Vision (2024)

    2024

  36. [36]

    In: Computer Vision and Pattern Recognition Workshop

    Wong, R., Camgoz, N.C., Bowden, R.: Learnt contrastive concept embeddings for sign recognition. In: Computer Vision and Pattern Recognition Workshop. pp. 1945–1954 (2023) SegSLR for Isolated Sign Language Recognition 15

    1945

  37. [37]

    arXiv preprint arXiv:2304.12620 (2023)

    Wu, J., Ji, W., Liu, Y., Fu, H., Xu, M., Xu, Y., Jin, Y.: Medical SAM adapter: Adapting segment anything model for medical image segmentation. arXiv preprint arXiv:2304.12620 (2023)

    Pith/arXiv arXiv 2023

  38. [38]

    In: German Conference on Pattern Recognition (2007)

    Zach, C., Pock, T., Bischof, H.: A duality based approach for realtime tv-l1 optical flow. In: German Conference on Pattern Recognition (2007)

    2007

  39. [39]

    In: International Conference on Computer Vision Workshop (2017)

    Zhang, L., Zhu, G., Shen, P., Song, J., Afaq Shah, S., Bennamoun, M.: Learning spatiotemporalfeaturesusing3dcnnandconvolutionallstmforgesturerecognition. In: International Conference on Computer Vision Workshop (2017)

    2017

  40. [40]

    In: AAAI Conference on Artificial Intelligence (2023)

    Zhao, W., Hu, H., Zhou, W., Shi, J., Li, H.: BEST: BERT pre-training for sign language recognition with coupling tokenization. In: AAAI Conference on Artificial Intelligence (2023)

    2023

  41. [41]

    IEEE Access5(2017)

    Zhu, G., Zhang, L., Shen, P., Song, J.: Multimodal gesture recognition using 3-d convolution and convolutional LSTM. IEEE Access5(2017)

    2017

  42. [42]

    In: Computer Vision and Pattern Recognition (2023)

    Zuo, R., Wei, F., Mak, B.: Natural language-assisted sign language recognition. In: Computer Vision and Pattern Recognition (2023)

    2023